A structure damage identification method based on modal strain energy and support vector machine (SVM) is put forward. The method is sensitive to structural damage location and has a strong anti-noise ability. With the strain energy index as input vector, support vector classification and regression machine are used to detect the damage location and damage degree, respectively. The simulated studies on damage identification are performed for a practical concrete girder bridge. The research results show that the presented approach has a higher accuracy in comparison with the traditional modal parameters based detection method. Moreover, it is indicated from the damage identification results using the noise-contaminated sample that the method possesses stronger anti-noise capability and stability than the method based on modal curvature and SVM.